Comparison and validation of outputs from global hydrological models against local hydrological data in South Africa

Study aim: To test the performance of 7 global hydrological models against local South African datasets from the Earth2Observe platform in terms of the monthly water balance estimation.

Methodology: The remotely-sensed global forcing datasets tested are actual evapotranspiration (ET) and runoff provided by Earth2Observe models. The local datasets were retrieved from the Water Resources of South Africa (WR2012) database, which quantifies the water resources of South Africa, Lesotho and Swaziland per quaternary catchment. The Quaternary Catchments consist of the lowest and most detailed level of operational catchment in the Department of Water and Sanitation (DWS), South Africa. An assessment of outputs between WR2012 and the ensemble outputs available to download from the seven Earth2Observe models (CEH, CNRS, UNIVK, UNIVU, METFR, JRC, and ECMWF) was conducted in South Africa. The mean annual runoff (MAR) time series analysis for the WR2012 and the Earth2Observe models were calculated. The Percentage of Bias (PoB) of each of the seven models with respect to the WR2012 data were computed as: %bias = (MAR_i - MAR_WR) x 100/ MAR_WR.

Results: Models JRC, UNIVU, and CNRS showed excessive bias values in large areas while the remaining four models generally exhibited large negative bias values, but with high positive values in the extreme Northeast and the dry areas of the country. The wetter regions of the country (Eastern Cape and KwaZulu Natal) displayed consistent results across all models. Apart from CNRS, all the results have an approximate water balance (precipitation = evapotranspiration + runoff). This is largely due to compensating over or under estimation of both precipitation and evapotranspiration.

Conclusion and Recommendations: The seven models are therefore rendered unreliable for water balance information in South Africa. Additionally, this study shows that ET data from the Earth2Observe platform is reliable to use, but the global runoff datasets should not be used in South Africa. It is recommended that organisations responsible for the generation of these global datasets issue warnings that these outputs have not been validated thus cannot be relied on.